Effects of attention and confidence on the hypothesized ERP correlates of recollection and familiarity

Transcription

1 Neuropsychologia 42 (2004) Effects of attention and confidence on the hypothesized ERP correlates of recollection and familiarity Tim Curran Department of Psychology, University of Colorado, Campus Box 345, Boulder, CO , USA Received 3 January 2003; received in revised form 12 November 2003; accepted 3 December 2003 Abstract Dual-process theories suggest that recognition memory is determined by two separate processes: familiarity and recollection. Experiment 1 behaviorally replicated past studies using the remember/know procedure to indicate that the amount of attention devoted to study influences both recollection and familiarity, but recollection more strongly. Experiments 1 and 2 assessed the effects of attention on two ERP components that have been hypothesized to be related to familiarity (FN400 old/new effect, ms, anterior) and recollection (parietal old/new effect, ms, posterior). Parietal old/new effects were reduced by divided attention, but FN400 old/new effects were not. Parietal ERPs ( ms) in experiment 2 increased with confidence in recognizing old items, but not new items. These results support the hypothesis that the parietal old/new effect is related to recollection Elsevier Ltd. All rights reserved. Keywords: Memory; Event-related potentials; Dual-process; Recollection; Familiarity Dual-process theories suggest that recognition memory Mecklinger, 2000; Nessler, Mecklinger, & Penney, 2001; performance is determined by separate processes of famil- Rugg, Mark et al., 1998; Tsivilis, Otten, & Rugg, 2001). iarity and recollection (Brainerd, Reyna, & Kneer, 1995; The present experiments further examine the utility of ERPs Hintzman & Curran, 1994; Jacoby, 1991; Mandler, 1980; for measuring activity related to recollection and familiarity. Norman & O Reilly, 2003; Reder et al., 2000; Yonelinas, The work of several groups (reviewed in Friedman & 1994). The dual-process perspective is theoretically attrac- Johnson, 2000; Mecklinger, 2000) has promoted the idea tive because it potentially explains various phenomena that that an early ( ms), mid-frontal, negative ERP effect have been challenging to single-process, familiarity-based is related to familiarity (here called the FN400 old/new efmodels (Clark & Gronlund, 1996; Hintzman & Curran, fect ; Curran, 1999, 2000; Curran & Cleary, 2003; Curran & 1994; Norman & O Reilly, 2003; Reder et al., 2000; Dien, 2003; Curran et al., 2002) 1, and a later ( ms), Yonelinas, 1997). However, the measurement methods used parietal, positive ERP effect is related to recollection (here to behaviorally dissociate familiarity and recollection effects called the parietal old/new effect ; Allan, Wilding, & have been controversial (reviewed by Humphreys, Dennis, Rugg, 1998; Curran, 1999, 2000; Curran & Cleary, 2003; Chalmers, & Finnigan, 2000; Yonelinas, 2002). An alter- Curran & Dien, 2003; Curran, Schacter, Johnson, & Spinks, native measurement approach is to neurophysiologically dissociate recollection and familiarity (Rugg & Yonelinas, 1 The relationship between the FN400 and the N400 is unclear. We use 2003) with methods such as functional magnetic resonance the FN400 label because our ms old/new differences are more frontally distributed than the centro-parietal N400 recorded to semantic imaging (fmri) (Dobbins, Rice, Wagner, & Schacter, 2003; incongruity (reviewed by Kutas & Van Petten, 1988, 1994). Many studies Eldridge, Knowlton, Furmanski, Bookheimer, & Engel, have found ms old/new effects associated with the centro-parietal 2000; Henson, Cansino, Herron, Robb, & Rugg, 2003) or N400 (e.g., Bentin & McCarthy, 1994; Besson, Fischler, Boaz, & Raney, event-related brain potentials (ERP) (Curran, 1999, 2000; 1992; Finnigan, Humphreys, Dennis, & Geffen, 2002; Olichney et al., Curran & Cleary, 2003; Curran & Dien, 2003; Curran, 2000; Van Petten, Kutas, Kluender, Mitchiner, & McIsaac, 1991) whereas others have found it to be more frontal (Mecklinger, 2000; Nessler et al., Tanaka, & Weiskopf, 2002; Düzel, Vargha-Khadem, Heinze, 2001; Rugg, Mark et al., 1998; Tsivilis et al., 2001). Its frontal manifes- & Mishkin, 2001; Guillem, Bicu, & Debruille, 2001; tation elsewhere has been called the mid-frontal (Tsivilis et al., 2001) or medial frontal (Friedman & Johnson, 2000) old/new effect. We consider Tel.: ; fax: previous studies describing ms old/new effects, however labeled address: tcurran@psych.colorado.edu (T. Curran). or spatially distributed, to be potentially relevant to the FN /$ see front matter 2004 Elsevier Ltd. All rights reserved. doi: /j.neuropsychologia

2 T. Curran / Neuropsychologia 42 (2004) ; Curran et al., 2002) 2. Our own work has started with the theoretical perspective that familiarity is the product of a global matching process that represents the similarity between a test item and all studied information (e.g., Clark & Gronlund, 1996; Gillund & Shiffrin, 1984; Hintzman, 1988; Shiffrin & Steyvers, 1997). According to this perspective, the FN400 seems related to familiarity because it responds similarly to studied items and similar lures. Examples include studied words and plurality-reversed lures (Curran, 2000), studied pictures and orientation-reversed lures (Curran & Cleary, 2003), studied geometric figures and visually similar lures (Curran et al., 2002), and studied words and semantically similar lures (Nessler et al., 2001). We have further assumed that recollection involves the retrieval of qualitatively specific information about individual items (e.g., Hintzman & Curran, 1994; Norman & O Reilly, 2003; Yonelinas, 1994). The observation that parietal old/new differences can be observed between studied and similar lures in the aforementioned experiments is consistent with a relationship to recollection (Curran, 2000; Curran & Cleary, 2003; Curran et al., 2002; Nessler et al., 2001). Other research has challenged the hypothesis that the FN400 is related to familiarity. For example, Olichney and colleagues (Olichney et al., 2000) found that the magnitude of ms parietal old/new effects (labeled LPC old/new effects ) was correlated with memory ability in amnesic patients and control subjects, and the parietal old/new differences were not significant when the amnesic group was considered as a whole. However, the N400 old/new effect was uncorrelated with memory performance (including recognition memory) and was normal in the amnesic patients. Thus, the authors conclude that the N400 old/new difference is unlikely to be related to the episodic memory processes such as familiarity. Second, Tsivilis et al. (2001) studied memory for objects under conditions in which the background context was incidentally varied. If both parts (object and context) of the display were present at study and test, the FN400 (labeled mid-frontal ms effect ) differed from conditions in which both the object and context were new. However, changing one part of the display from study to test (either the object or the context, but not both) abolished the FN400 old/new effect. In other words, the FN400 was insensitive when only part of the display was familiar. Thus, Tsivilis et al. suggest that the FN400 old/new differences may be related to a noveltydetection process that is correlated with, but downstream to, familiarity assessment. The hypothesis that the parietal effect is related to recollection has been more widely accepted, but it has also been 2 The parietal old/new effect co-occurs with the P300 component (Bentin & McCarthy, 1994; Spencer, Vila Abad, & Donchin, 2000), and has been variously labeled the P300 old/new difference (Johnson, 1995), the late ERP old/new effect (Rugg, 1995), and the P600 old/new effect (Curran, 1999; Rugg & Doyle, 1992). questioned. Finnigan et al. (2002) found that a ms N400 effect recorded over the parietal scalp varied with presentation frequency whereas a later ms parietal effect varied with the accuracy of recognition judgments. They suggested these two effects may respectively be related to strength and decision processes underlying a single-process, signal-detection-like recognition memory mechanism, rather than being related to separate familiarity and recollection processes. Finnigan et al. s meaning of strength and the present meaning of familiarity are both inspired by the global matching models of memory, so we are in agreement on the nature of the N400 old/new effect (although their N400 was parietal rather than frontal). According to Finnigan et al. s perspective, however, parietal old/new effects could simply reflect decision-related processes such as confidence or criterion setting rather than indexing a secondary recollection process. These differing perspectives on the functional interpretation of the FN400 and parietal old/new effects suggest that further research is needed. Another potentially fruitful approach for linking the FN400 effect to familiarity and the parietal effect to recollection is to establish a correspondence between these ERP effects and results from behavioral measurement methods. For example, several studies have measured ERPs during the remember/know paradigm (for methodological review of the remember/know procedure see, Gardiner & Richardson-Klavehn, 2000). The parietal old/new effect is larger when recognition is based upon a subjective experience of remembering than knowing (Düzel, Yonelinas, Mangun, Heinze, & Tulving, 1997; Rugg, Schloerscheidt, & Mark, 1998; Smith, 1993; Trott, Friedman, Ritter, Fabiani, & Snodgrass, 1999), though these amplitude differences might be attributable to greater latency jitter in knowing trials (Spencer et al., 2000). ERP correlates of knowing possibly related to familiarity were less clear in these experiments, although an early knowing-related frontotemporal effect has been observed that may be related to the FN400 familiarity effect (Düzel et al., 1997). Other experiments have taken an indirect approach by showing that variables affecting familiarity and recollection estimates in previous studies have corresponding effects on ERPs. For example, the parietal old/new effect has been observed to be larger for words than pseudowords, but the FN400 old/new effect was similar for words and pseudowords (Curran, 1999). The parietal effects are consistent with findings that remembering is greater for words than pseudowords (Curran, Schacter, Norman, & Galluccio, 1997; Gardiner & Java, 1990; Whittlesea & Williams, 2000). However, estimates of familiarity derived from knowing from these same experiments with words and pseudowords have been less consistent, so the correspondence between the FN400 effects and familiarity is less certain. Some of these differences may be attributable to characteristics of the pseudowords (Whittlesea & Williams, 2000).

3 1090 T. Curran / Neuropsychologia 42 (2004) Comparisons between behavioral and electrophysiological indices of recollection and familiarity would be stronger in cases where existing behavioral work is more clear-cut. Yonelinas (2001) has recently shown that familiarity and recollection estimates can be similar across the process-dissociation, remember/know, and receiver operating characteristics (ROCs) procedures. For example, all three procedures showed that both recollection and familiarity are greater following semantic processing at encoding rather than perceptual processing (i.e., levels of processing, LOP, see also Gardiner, 1988; Jacoby, 1991; Rajaram, 1993; Toth, 1996; Wagner, Gabrieli, & Verfaellie, 1997). Several ERP studies have shown that the parietal old/new effect increases with LOP (Paller & Kutas, 1992; Paller, Kutas, & McIsaac, 1995; Rugg, Allan, & Birch, 2000; Rugg, Mark et al. 1998). Among those studies that considered LOP effects on the FN400, the FN400 was unaffected by LOP when conditions were randomized within each block (Rugg, Mark et al., 1998), but the FN400 old/new effect was observed after deep (but not shallow) study when conditions were manipulated between blocks (Rugg et al., 2000). Yonelinas (2001) behavioral studies used an intermediate design in which study conditions were blocked, but followed by a mixed test list. Thus, although some behavioral consistency has been demonstrated, the FN400 results have been less reliable. Yonelinas (2001) also manipulated divided attention at study. Both ROC and remember/know experiments indicated that divided attention affected recollection and familiarity, but recollection more strongly (see also, Gardiner & Parkin, 1990; Jacoby & Kelley, 1991; Parkin, Gardiner, & Rosser, 1995; Reinitz, Morrissey, & Demb, 1994). One study has investigated effects of divided attention at study on ERPs recorded during study (Mangels, Picton, & Craik, 2001), but none have investigated how divided attention at study affects the putative ERP correlates of familiarity and recollection during retrieval. The present experiments manipulated attention during study and measured ERPs during recognition tests to determine if the FN400 and parietal old/new effects conformed to behavioral indications that both familiarity and recollection are reduced by dividing attention. 1. Experiment Method Subjects Fifty-one students from Case Western Reserve University participated in one session (approximately 2 h) in partial fulfillment of a course requirement. Data from 12 subjects were discarded because of excessively high sensor impedances (five), excessive eye blinking (four), excessive 3 Experiment 1 was run after experiment 2 but their presentation is reversed to facilitate exposition. movement (one), computer malfunction, or mis-instruction (one). Twenty-eight of the 39 subjects retained for analyses were male. The behavioral results of the included and excluded subjects were qualitatively similar Stimuli Stimuli consisted of 300 common English words. Words were centrally presented in white against a black background. Visual display was a 15 in. Apple multiscan color monitor. The words were divided into four sets of 75 that were roughly equated for word frequency (MN = 15.39, S.D. = 18.86, range = 0 : 99, (Kucera & Francis, 1967)) and number of letters (MN = 5.42, S.D. = 1.04, range = 4 : 7). These sets were counterbalanced through the attention and old/new conditions across subjects. Additional words with similar characteristics were used as practice and buffer items. Stimuli for the divided attention task were digital recordings of a single male voice speaking the digits from0to Design Attention during study (full, divided) and memory status (old, new) of the words were manipulated within subjects. Each of four experimental blocks included two study lists (25 words per list) followed by a single test list (75 words). Each pair of study lists included one full-attention list and one divided-attention list. The order of full/divided lists was alternated across blocks and counterbalanced across subjects. Test lists included randomly intermixed words from the full, divided, and new conditions (25 per condition per block) Procedure Subjects were instructed before completing a short practice block that consisted of two 25-word study lists followed by a single 12-word test list. A Geodesic Sensor Net was applied between the practice and experimental blocks. Subjects studied two word lists (one full and one divided) before each recognition test. Each study list contained 25 words surrounded by one-word primacy and recency buffers. Each study trial included a 600 ms fixation sign (+) preceding a 1000 ms word. Subjects were instructed to study each word for an upcoming recognition test. Because subjects were not given an explicit encoding task, it is likely that the two conditions differ in level of processing in addition to attention. During divided-attention lists, subjects additionally heard a random digit presented at the beginning of each trial. Subjects were instructed to press a key each time three odd digits occurred consecutively (following Craik, 1982; Jacoby & Kelley, 1991). No digits were presented during full attention lists. A rest period of at least 2 min intervened between each study and test list. Each test block contained 75 test words (25 per condition) that were divided into five sub-blocks, so that subjects could periodically rest their eyes. Each sub-block began with a non-studied word that was not included in analyses

4 T. Curran / Neuropsychologia 42 (2004) to minimize post-break movement artifacts. All test-trial event onsets were synchronized to the refresh cycle of the monitor. Trials began with a variable duration plus sign ( ms), followed by a 2000 ms test-word presentation, followed by a question mark. Subjects were instructed to withhold their responses until the question mark appeared, and to minimize blinking and other movements. Test trials were concluded upon the subject s response. Subjects responded by pressing remember, know or new keys. Assignment of response categories to keys/hands was counterbalanced across subjects. Remember and know keys were always assigned to the first two fingers of one hand and the new key was always assigned to the first finger of the other hand. Subjects responded remember when they could recollect specific details about the word from the study episode, but know when they judged the word to be studied without the recollection of details EEG/ERP methods Scalp voltages were collected with a 128-channel Geodesic Sensor Net TM (Tucker, 1993) connected to an ac-coupled, 128-channel, high-input impedance amplifier (200 M, Net Amps TM, Electrical Geodesics Inc., Eugene, OR). Amplified analog voltages ( Hz bandpass, 3 db) were digitized at 250 Hz. Individual sensors were adjusted until impedances were less than 50 k. The EEG was collected continuously during test blocks, digitally low-pass filtered at 40 Hz, and then segmented into epochs starting 100 ms before test-word onset and lasting 1500 ms after. Trials were discarded from analyses if they contained incorrect responses, eye movements (EOG over 70 V), or more than 20% of channels were bad (average amplitude over 200 V or transit amplitude over 100 ms). Individual bad channels were replaced on a trial-by-trial basis with a spherical spline algorithm (Srinivasan, Nunez, Silberstein, Tucker, & Cadusch, 1996). Consistently bad channels for a given subject were replaced throughout that subject s entire dataset (bad channels per subject: median = mode = 1, range = 0 3). EEG was measured with respect to a vertex reference (Cz), but an average-reference transformation was used to minimize the effects of reference-site activity and accurately estimate the scalp topography of the measured electrical fields (Bertrand, Perin, & Pernier, 1985; Curran, Tucker, Kutas, & Posner, 1993; Dien, 1998; Lehman & Skrandies, 1985; Picton, Lins, & Scherg, 1995; Tucker, Liotti, Potts, Russell, & Posner, 1994). Average-reference ERPs are computed for each channel as the voltage difference between that channel and the average of all channels. The average reference was corrected for the polar average 4 Remember/know instructions were standard except that recollect and familiar labels were used rather than remember and know because the former terms tend to cause less confusion (Norman, 2002). The judgments are described herein as remember and know to clearly differentiate between subjects responses (remember, know) and the underlying processes of interest (recollection, familiarity). I thank Ken Norman for providing instructions from his experiments. reference effect (Junghöfer, Elbert, Tucker, & Braun, 1999). ERPs were baseline-corrected with respect to a 100 ms prestimulus recording interval. 2. Results 2.1. Analysis strategy Only 14 subjects had a sufficient number (>15) of correct, artifact-free remember and know trials in both the full and divided conditions. More subjects could be retained by comparing remember and know trials collapsed across full/divided attention (n = 35), or by comparing full and divided trials collapsed across remember/know (n = 39). ERP analyses from the remember/know and divided/full attention comparisons are presented separately below. Results from the 14 subjects with sufficient trials in all remember/know full/divided attention categories are not presented because they were not consistent with the results of the analyses from the larger samples. 5 Behavioral results are presented for the larger of the two samples (n = 39) Behavioral results Table 1 shows the mean proportion of remember, know, and new responses given in each condition. As detailed by Yonelinas (2002), recollection processes (R) can be directly estimated from the proportion of remember responses, but familiarity processes (F) cannot be directly estimated from the proportion of know responses because the latter judgment is contingent on recollection failure. Familiarity can be estimated by assuming that familiarity and recollection are independent, and dividing the know proportion by 1 minus the recollect proportion (F = K/[1 R]). Because we are particularly interested in estimating recollection (R) and familiarity (F) in the present experiment, analyses focused on these estimates. 6 An estimate type (R, F) by condition (full, divided, new) repeated measures ANOVA resulted in both main effects and the interaction being highly significant (all F>19, all P<0.001). Within each estimate type, contrasts indicated that each condition significantly differed from each other condition (all P<0.001), so divided attention reliably affected both recollection and familiarity. The interaction suggests that divided attention reduced recollection more than 5 Remember/know full/divided ANOVAs on the FN400 and parietal effects (as described later) from these 14 subjects revealed only a significant effect of full/divided on the FN400. This is an effect that was not significant when full and divided conditions were compared across 39 subjects. Furthermore, these analyses failed to replicate the remember/know and full/divided effects that were significant in the subsequent analyses. 6 The observation that remember responses were given to 9% of new items may indicate that some subjects misunderstood instructions, assuming that false recollection rarely occurs with unrelated lures. However, the pattern of results did not qualitatively differ when subjects with high false remember rates (>10%) were omitted.

5 1092 T. Curran / Neuropsychologia 42 (2004) Fig. 1. Grand average ERPs from experiment 1. Shown channels are representative of the international system (Jasper, 1958). Channels are labeled according to Geodesic Electrode Net numbers (see Fig. 2) along with their nearest equivalent location. Table 1 Behavioral results from experiment 1 Full Divided New Remember (R) (recollection) Know (K) New (N) Familiarity (F) (=K/[1 R]) Notes: Top three rows are the proportions of R, K, and N responses within each condition. familiarity. Overall, these results agree with others reviewed in the introduction Remember/know ERP results Correct trials given remember or know responses were collapsed across the full and divided attention conditions, and compared with correctly rejected new trials. Thirty-five subjects had at least 19 acceptable trials in each condition (mean trials/condition: remember = 81, know = 50, new = 54). ERPs recorded near selected locations from the international system are shown in Fig. 1 (Jasper, 1958). FN400 and parietal old/new effects were analyzed within temporal windows (FN400: ms; parietal: 7 Recollection results were very similar between the subjects who were included (n = 39) versus excluded from the primary analyses. Familiarity estimates were qualitatively similar, but lower for excluded subjects: full = 0.47 divided = 0.34, new = 0.20 (compare to values from included subjects in Table 1) ms) and spatial regions (described below) identified in previous studies (following, Curran, 2000; Curran & Cleary, 2003; Curran & Dien, 2003; Curran et al., 2001, 2002). Condition (remember, know, new) hemisphere repeated-measures ANOVAs were computed within each window separately. These were followed by planned condition hemisphere ANOVAs to specifically assess old/new effects associated with remember judgments (remember versus new), old/new effects associated with know judgments (know versus new), and differences between remembering and knowing. All condition main effects are reported, regardless of significance. Hemisphere condition interactions are reported only when significant. ANOVA results are summarized in Tables 2 and 3, adjusted according to the conservative Geisser-Greenhouse procedure for sphericity violations (Winer, 1971). Table 2 Experiment 1: FN400 ANOVA results ( ms, LAS and RAS regions) Effect d.f. F M.S.E. P Remember, know, new 2, <0.01 Remember vs. new 1, <0.01 Know vs. new 1, <0.05 Remember vs. know 1, n.s. Full, divided, new 2, <0.001 Full vs. new 1, <0.001 Div vs. new 1, <0.01 Full vs. div 1, n.s. Notes:d.f. = degree of freedom; div = divided; and n.s. = not significant.

6 T. Curran / Neuropsychologia 42 (2004) Table 3 Experiment 1: parietal ANOVA results ( ms, LPS and RPS regions) Effect d.f. F M.S.E. P Remember, know, new 2, <0.001 Hem 2, <0.001 Remember vs. new 1, <0.001 Hem 1, <0.001 Know vs. new 1, n.s. Hem 1, <0.01 Remember vs. know 1, <0.001 Hem 1, Full, div, new 2, <0.001 Hem 2, <0.001 Full vs. new 1, <0.001 Hem 1, <0.001 Div vs. new 1, <0.05 Hem 1, <0.001 Full vs. div 1, <0.01 Notes: d.f. = degree of freedom; div = divided; hem = hemisphere; and n.s. = not significant Remember/know: FN400 results ( ms) The FN400 was analyzed over two anterior, superior channels groups centered near the standard F3 and F4 locations. The regions are labeled left and right anterior/superior (LAS and RAS, see Fig. 2) and ERPs averaged within these regions are plotted in Fig. 3. Mean LAS and RAS amplitude Fig. 3. Grand average ERPs from experiment 1. ERPs are averaged across channels within the LAS, RAS, LPS, and RPS regions of interest shown in Fig. 2. between 300 and 500 ms was the dependent measure (Fig. 4). As shown in Table 2, significant old/new differences were observed for old items associated with both remember and know responses, and the remember and know categories did not differ. Thus, the FN400 old/new effect was found to be similar for both remembering and knowing Remember/know parietal results ( ms) Mean LPS and RPS (left and right posterior/superior) amplitude within each channel group, between 400 and 800 ms, was the dependent measure (see Fig. 2 for locations, Fig. 3 for ERP plots, Fig. 4 for mean amplitudes, and Table 3 for ANOVA results). A greater parietal positivity was observed for remembered words than known words, especially over the left hemisphere. Furthermore, the parietal effect was larger for remembering than knowing. The significant know/new hemisphere interaction suggested knowing was more positive than new over the left hemisphere (P = 0.08), but new was more positive than knowing over the right hemisphere (P <0.05). It should be acknowledged that condition differences in the amount of latency jitter across trials may contribute to the observed amplitude differences (Spencer et al., 2000). However, ERP conditions differing in latency Fig. 2. Approximate sensor locations and selected locations from the system. Channels within regions of interest used in ANOVAs are shown in black. L = left, R = right, A = anterior, P = posterior, S = superior, and I = inferior. Fig. 4. Mean amplitudes ( V) corresponding to the FN400 (a) and parietal (b) old/new effects in experiment 1. Error bars represent the standard error of the old/new difference (hence, the absence of error bars for the new conditions). (a) Mean amplitudes across the LAS and RAS regions from 300 to 500 ms. (b) Mean amplitudes within the LPS and RPS regions from 400 to 800 ms.

7 1094 T. Curran / Neuropsychologia 42 (2004) Fig. 5. Topography of the old/new differences estimated by spherical-spline interpolations (Srinivasan et al., 1996). The front of the head is depicted at the top of each oval. jitter often differ in the peakedness of the waveforms, such that low jitter leads to peaked waveforms and high jitter leads to a flatter, shallower waveform. The present ERP conditions do not clearly differ in peakedness (see Fig. 3) Remember/know: topographic comparisons Further analyses investigated topographic differences between the FN400 and parietal old/new effects (see Fig. 5). Because the temporal overlap between these effects would obscure topographic differentiation, the ms parietal window was limited to ms to maintain separation from the ms FN400 window. Old/new differences within each temporal window were computed separately from the full and divided attention conditions, and the mean of these differences was taken within each of the eight spatial regions shown in Fig. 2 (Curran, 1999, 2000; Curran & Cleary, 2003; Curran et al., 2002). Amplitude differences between the conditions and temporal windows were removed by vector-length normalization (McCarthy & Wood, 1985). The normalized differences were the dependent measures in a condition (remember, know) time ( , ) hemisphere anterior/posterior superior/inferior repeated measures ANOVA. Interactions between time and the topographic factors would indicate that the FN400 and parietal old/new effects are topographically different. As seen in Fig. 5, both the FN400 ( ) and parietal ( ) old/new differences are positive-going over superior regions and negative-going over inferior regions (Curran, 1999, 2000; Curran & Cleary, 2003; Curran et al., 2002). The time hemisphere anterior/posterior inferior/superior interaction indicated that, relative to the FN400 old/new effect, the parietal effect had larger left posterior/superior (i.e., left parietal) differences and opposite-going right anterior/inferior differences, F(1, 34) = 4.85, M.S.E. = 0.01, P<0.05. This overall topographic difference between the FN400 and parietal old/new effects was qualified by several interactions between remember/know, time, and topography. One generally consistent interpretation of the following interactions is that remembering and knowing were associated with similar ms old/new effects, but ms old/new effects were larger and topographically different for remembering than knowing. The time remember/know anterior/posterior interaction suggested that anterior differences were more positive than posterior differences for all time remember/know combinations except the ms differences related to remembering; in this case, posterior differences were more positive than anterior, F(1, 34) = 10.52, M.S.E. = 0.05, P<0.01. The time remember/know inferior/superior interaction suggested that ms FN400 effects showed similar superior > inferior differences for knowing and remembering, but the ms parietal effects showed much larger superior > inferior differences related to remembering than knowing. The time remember/know inferior/superior hemisphere interaction can be interpreted similarly to the previous interaction, with the additional hemispheric interaction capturing the fact that the large superior > inferior difference associated with remembering from ms especially involved left, superior and right, inferior regions. In summary, the time region interactions suggest that the FN400 and parietal old/new effects are topographically different. Further interactions suggest that remembering and knowing primarily differ at the parietal old/new effect Remember/know: late frontal ( ms) Although the experiment was not designed with late frontal effects in mind, they were examined for completeness because old > new differences are commonly observed

8 T. Curran / Neuropsychologia 42 (2004) Fig. 6. Grand average ERPs from experiment 1. Shown channels are representative of the international system (Jasper, 1958). Channels are labeled according to Geodesic Electrode Net numbers (see Fig. 2) along with their nearest equivalent location. (Allan et al., 1998; Curran & Friedman, 2003; Curran et al., 2001; Johnson, Kounios, & Nolde, 1996; Ranganath & Paller, 2000; Wilding, 1999; Wilding & Rugg, 1997a,b). Six channel groups (seven channels/group) were defined near the Fp1/Fp2, F7/F8, and F3/F4 locations (following Curran & Friedman, 2003). 8 Mean amplitude from 1000 to 1500 ms was entered into a condition (remember, know, new) hemisphere frontal region (Fp1/2+, F7/8+, F3/4+) ANOVA. The only significant condition effect was a condition by hemisphere interaction, suggesting that condition differences were only observed over the right hemisphere, F(1, 34) = 4.50, M.S.E. = 1.09, P< Comparing conditions within the right hemisphere suggested that both old/new differences were significant (remember > new, P < 0.001; know > new, P < 0.05), but the remember and know conditions did not significantly differ (see RAS region of Fig. 3). These results are consistent with previous ERP remember/know experiments that have examined late frontal effects (Düzel et al., 1997; Rugg, Schloerscheidt et al., 1998; Smith, 1993; Trott et al., 1999) Full/divided attention ERP results Correct trials within the full and divided conditions were collapsed across remember and know responses and compared with correctly rejected new trials. Thirty-nine subjects had at least 19 acceptable trials in each condition (mean trials/condition: full = 71, divided = 56, new = 52). ERPs recorded near selected locations from the international system are shown in Fig. 6 (Jasper, 1958). Analyses were analogous to those previously comparing remember, know, and new conditions. ERP waveforms from regions of interest are show in Fig. 7, mean amplitudes in Fig. 8, and ANOVA results in Table 1 (FN400) and Table 2 (parietal). 8 The channels falling within each region were: Fp1+ (18, 19, 22, 23, 24, 26, 27), Fp2+ (2, 3, 8, 9, 10, 14, 15), F7+ (F7, 33, 39, 40, 44, 45, 128), F8+ (F8, 1, 115, 116, 120, 121, 125), F3+(12, 13, 20, 21, 25, 29, 30), F4+(4, 5, 112, 113, 118, 119, 124), see Fig. 2 for channel locations. Note that the F3+ and F4+ regions are identical to those previously labeled LAS and RAS. Fig. 7. Grand average ERPs from experiment 1. ERPs are averaged across channels within the LAS, RAS, LPS, and RPS regions of interest shown in Fig. 2.

9 1096 T. Curran / Neuropsychologia 42 (2004) Fig. 8. Mean amplitudes ( V) corresponding to the FN400 (a) and Parietal (b) old/new effects in experiment 1. Error bars represent the standard error of the old/new difference (hence, the absence of error bars for the new conditions). (a) Mean amplitudes across the LAS and RAS regions from 300 to 500 ms. (b) Mean amplitudes within the LPS and RPS regions from 400 to 800 ms Full/divided attention: FN400 results ( ms) Mean LAS and RAS amplitude between 300 and 500 ms was the dependent measure. As shown in Table 2, significant old/new differences were observed in both the full and divided conditions, and the full and divided conditions did not differ. Thus, the FN400 old/new effect was found to be of similar magnitudes following either full or divided attention at study Full/divided attention: parietal results ( ms) Mean LPS and RPS amplitude between 400 and 800 ms was the dependent measure. As shown in Table 3, significant old/new differences were observed in both the full and divided conditions, but the effect was greater following full than divided attention. Each of the old/new differences was greater over the left than right hemisphere, as is often observed (Allan et al., 1998). Thus, the parietal old/new difference was greater for words studied with full attention than for words studied with divided attention Full/divided attention: topographic comparisons Normalized old/new differences (Fig. 5) were the dependent measures in a condition (full, divided) time ( , ms) hemisphere anterior/posterior superior/inferior repeated measures ANOVA. The time anterior/posterior interaction indicated that ms FN400 differences were larger anteriorly whereas the ms parietal differences were larger posteriorly, F(1, 38) = 4.54, M.S.E. = 0.14, P<0.05. This interaction upholds the a priori choice of anterior regions for FN400 analyses and posterior regions for the parietal analyses reported earlier. Several other lower-order topographic interactions were observed, but interpretation hinges on two higher-order interactions, so only these are reported. The time hemisphere anterior/posterior superior/inferior was significant, F(1, 38) = 4.43, M.S.E. = 0.01, P<0.05. The four-way interaction stems from the fact that, relative to the FN400 old/new effect, the parietal effect is associated with larger left posterior/superior (i.e., left parietal) differences and opposite-going right anterior/inferior differences. A similar observation is captured by the significant condition time hemisphere superior/inferior interaction, F(1, 38) = 8.76, M.S.E. = 0.01, P<0.01. This interaction indicates that the early and late old/new differences had similar topographic patterns for the divided-attention condition, but for the full-attention condition the opposite going left/ superior and right/inferior differences were larger in the later window. In summary, the time region interactions suggest that the FN400 and parietal old/new effects are topographically different. The last interaction with condition reiterates the results of the prior analyses indicating that dividing attention affected the later parietal old/new effect but not the earlier FN400 old/new effect Full/divided attention: late frontal effects ( ) A condition (full, divided, new) 2 hemisphere 3 frontal regions (Fp1/2+, F7/8+, F3/4+) ANOVA revealed a significant condition hemisphere interaction, F(2, 76) = 3.81, M.S.E. = 1.10, P<0.05. Right hemisphere voltages were more positive for old than new items, as is typically observed (see RAS region of Fig. 7). Contrasts (M.S.E. = 1.10) focused on right, frontal regions revealed that full (MN = 1.74 V, F(1, 76) = 7.48, P = 0.01) and divided (MN = 1.93 V, F(1, 76) = 17.37, P<0.001) conditions were more positive than new (MN = 1.38 V), but full and divided did not significantly differ (F(1, 76) = 1.87). 3. Experiment 2 Experiment 1 used both behavioral and ERP methods to examine the effects of divided attention on recollection and familiarity. Familiarity and recollection were estimated from behavioral remember/know judgments by assuming the two processes are independent (Yonelinas, 2002). FN400 ERP old/new effects were hypothesized to be related to familiarity whereas parietal ERP old/new effects were hypothesized to be related to recollection (Curran, 1999, 2000; Curran & Cleary, 2003; Curran & Dien, 2003; Curran et al., 2002; Mecklinger, 2000; Nessler et al., 2001; Rugg, Mark et al., 1998). This hypothesis was upheld by the finding that parietal old/new effects were larger when associated with remembering than knowing, but remembering and knowing produced similar FN400 old/new effects. Both the behavioral estimates and ERP indices suggested that divided attention during study impairs recollection more than

10 T. Curran / Neuropsychologia 42 (2004) familiarity. At a more detailed level, however, the behavioral remember/know estimates and ERP analyses suggest different effects of divided attention on familiarity. In accord with previous behavioral studies, the remember/know results suggest that familiarity is hurt by divided attention, whereas ERP analyses suggest that the hypothesized FN400 old/new effect was unaffected by dividing attention. The goals of experiment 2 were to replicate experiment 1 and to investigate the effects of confidence. Understanding how confidence influences the parietal old/new effect is important for assessing its functional significance. Words studied with full attention are recognized with greater confidence than those studied with divided attention (Yonelinas, 2001). Some dual-process models conceptualize recollection as a high-threshold process that leads to high-confidence responses (Norman & O Reilly, 2003; Yonelinas, 1994, 2001). From this perspective, conditions associated with higher recollection rates (e.g., full attention) would naturally be associated with higher confidence. From a single-process perspective that denies the existence of separate familiarity and recollection processes, confidence differences may be considered to reflect processes related more to decision making than to memory retrieval per se (Finnigan et al., 2002). Confidence ratings were collected in experiment 2. Both the single-process and dual-process perspectives predict that the parietal old/new effect should be affected by confidence in recognizing old items, but they differ with respect to predicted effects of confidence on new items. If the parietal old/new effect is related to a high-threshold recollection process, it should be affected by confidence in recognizing old items, but not by confidence in rejecting new items. This follows from the high-threshold assumption that new items are very rarely recollected. If the parietal old/new effect is related to decision making or criterion setting, it should be affected by confidence in both old and new items because confidence differences in either case reflect distance from the response criterion. 4. Method The method was identical to experiment 1, except that recognition responses were made on a four-point confidence scale: 1 = sure new, 2 = maybe new, 3 = maybe old, and 4 = sureold (order reversed for half the subjects). Subjects were 58 students from Case Western Reserve University who participated in partial fulfillment of a course requirement. Data from 21 subjects were discarded because of excessive eye blinking (eight), excessively high sensor impedances (four), computer malfunction (four), amplifier malfunction (two), incorrect sampling rate (two), or itching from saline solution (one). Thirty of the 37 subjects retained for analyses were male. The behavioral results of the included and excluded subjects were qualitatively similar. 5. Results 5.1. Behavioral results Table 4 shows recognition performance for subgroups of subjects used in particular ERP analyses described below. Only subjects with at least 20 artifact-free trials in each condition were included in each analysis. All good subjects were those with sufficient trials/condition, regardless of confidence. HC old includes all subjects with sufficient high-confidence hits in both the full and divided conditions. HC/LC divided subjects had sufficient high confidence and low confidence response in the divided attention condition. HC/LC new subjects had sufficient high confidence and low confidence response in the new condition. Focusing on all good subjects, as in experiment 1, hit rates were higher following full than divided attention, t(36) = 12.53, S.E. = 0.02, P< (first row of Table 4). When compared with the false alarm rate to new items (29%), each condition showed above chance discrimination between old and new: full, t(36) = 19.43, S.E. = 0.03, P<0.0001; divided, t(36) = 16.31, S.E. = 0.02, P< Table 4 also shows the proportion of accurate trials on which subjects gave the most confident ( sure ) response. All pairwise differences were significant such that confidence in correct responses was ranked full > divided > new (all P<0.01). Notably, with only one exception, the pattern of statistical effects reported for all good subjects was replicated in each of the subgroups. The one exception was the HC/LC new group which did not show a significant confidence difference between divided and new conditions ERP results Analyses ideally would include separate ERPs from highand low-confidence responses in each condition (correct full, divided, and new trials). However, only three subjects yielded at least 20 good trials in each of the six categories (two confidence three condition). Confidence-related effects were assessed in the following analyses by examining subsets of subjects with sufficient trials in particular conditions. All subjects retained in each analysis had at least 21 acceptable trials in each condition. Table 4 Behavioral results from experiment 2 Group n Accuracy Confidence Full Divided New Full Divided New All good HC old HC/LC divided HC/LC new Notes: Confidence refers to the proportion of accurate trials given highconfidence responses. HC = high confidence; and LC = low confidence.

11 1098 T. Curran / Neuropsychologia 42 (2004) Fig. 9. Grand average ERPs from all good subjects from experiment 2. Shown channels are representative of the international system (Jasper, 1958). Channels are labeled according to Geodesic Electrode Net numbers (see Fig. 2) along with their nearest equivalent location All good subjects (across confidence) Initial analyses included all subjects with sufficient accurate trials in the full, divided, and new conditions regardless of confidence (trials/condition: full = 69, divided = 51, new = 60). These analyses were meant to replicate those in experiment 1. Corresponding ERP plots are shown in Fig. 9 (10 20 locations) and Fig. 10 (regions of interest). Mean amplitudes are shown in Fig. 11. The FN400 ( ms, LAS and RAS regions) was significantly more negative for new than divided condition (Table 5, Fig. 11). Full/new Fig. 10. Grand average ERPs from all good subjects from experiment 2. ERPs are averaged across channels within the LAS, RAS, LPS, and RPS regions of interest shown in Fig. 2. Table 5 Experiment 2: FN400 ANOVA results ( ms, LAS and RAS regions) Effect d.f. F M.S.E. P All good Full, div, new 2, <0.01 Full vs. new 1, Div vs. new 1, <0.001 Full vs. div 1, HC old Full (HC), div (HC), new 2, <0.01 Full (HC) vs. new 1, <0.05 div (HC) vs. new 1, <0.001 Full (HC) vs. div (HC) 1, HC/LC div Div (HC), div (LC), new 2, Div (HC) vs. new 1, <0.05 Div (LC) vs. new 1, Div (HC) vs. div (LC) 1, n.s. HC/LC new Full, div, new (HC), new (LC) 3, <0.01 Full vs. new (HC) 1, <0.01 Div vs. new (HC) 1, <0.01 Full vs. new (LC) 1, n.s. Div vs. new (LC) 1, n.s. New (HC) vs. new (LC) 1, Notes: Exp = experiment; d.f. = degree of freedom; div = divided; LC = low confidence; HC = high confidence; and n.s. = not significant.

12 T. Curran / Neuropsychologia 42 (2004) Fig. 11. Mean amplitudes ( V) corresponding to the FN400 (a) and Parietal (b) old/new effects in experiment 2. Error bars represent the standard error of the old/new difference (hence, the absence of error bars for the new conditions). The dashed line (- - -) shows all good subjects, and the solid line ( ) shows the HC Old group. (a) Mean amplitudes across the LAS and RAS regions from 300 to 500 ms. (b) Mean amplitudes across the LPS and RPS regions from 400 to 800 ms. and full/divided differences were marginally significant. The parietal old/new effect ( ms, LPS and RPS regions) was significant following both study conditions, but was larger following full than divided attention (Table 6). Although statistical details varied, the overall pattern of results was qualitatively similar to experiment 1. Dividing attention reduced the parietal old/new effect, but not the FN400 old/new effect. Notably, the marginal trend toward full/divided FN400 differences was in the opposite direc- Table 6 Experiment 2: parietal ANOVA results ( ms, LPS and RPS regions) Effect d.f. F M.S.E. P All good Full, divided, new 2, <0.001 Full vs. new 1, <0.001 Div vs. new 1, Full vs. div 1, <0.01 HC old Full (HC), divided (HC), new 2, <0.001 Full (HC) vs. new 1, <0.001 Div (HC) vs. new 1, <0.01 Full (HC) vs. div (HC) 1, HC/LC divided Div (HC), div (LC), new 2, <0.05 Div (HC) vs. new 1, <0.05 Div (LC) vs. new 1, n.s. Div (HC) vs. div (LC) 1, <0.05 HC/LC new Full, div, new (HC), new (LC) 3, <0.001 Full vs. new (HC) 1, <0.001 Div vs. new (HC) 1, n.s. Full vs. new (LC) 1, <0.001 Div vs. new (LC) 1, New (HC) vs. new (LC) 1, n.s. Notes: Exp = experiment; d.f. = degree of freedom; div = divided; LC = low confidence; HC = high confidence; and n.s. = not significant. Fig. 12. Grand average ERPs from the HC old group of experiment 2. ERPs are averaged across channels within the LAS, RAS, LPS, and RPS regions of interest shown in Fig. 2. tion such that the FN400 old/new difference was larger following divided than full attention High confidence old (HC old) The next analysis focused on subjects with sufficient high confidence hits in both the full and divided attention conditions. These analyses were intended to assess the effects of divided attention on ERP old/new effects when confidence differences between conditions are minimized (e.g., Rugg & Doyle, 1992; Wilding & Rugg, 1996). Each of these conditions was compared with ERPs to all correct rejections (regardless of confidence) to examine old/new differences. The mean number of acceptable trials per condition was full = 58, divided = 32, new = 63. Corresponding waveforms are shown in Fig. 12. The FN400 old/new effect was significant following either divided or full attention, and the full/divided difference was marginal (Table 5). The parietal old/new also was significant following either divided or full attention with full being more positive than divided (Table 6). The overall patterns were quite similar, whether or not low-confidence trials were included (Fig. 11) High versus low confidence divided attention (HC/LC divided) The HC/LC divided group included only subjects with sufficient high- and low-confidence hits in the divided attention condition, so the effects of confidence could be directly assessed within a single condition. ERPs associated with hits also were compared with ERPs to all correct rejections (regardless of confidence). The mean number of acceptable trials per condition was high = 31, low = 27, new = 64. Corresponding ERP waveforms are shown in Fig. 13, and mean amplitudes in the high and low categories are shown with the dashed lines (- - -) in Fig. 14. The FN400 old/new effects was significant for high confidence hits, but not for low confidence hits (Table 5). The parietal old/new effect similarly was significant for only high-confidence hits, and a significant difference emerged when high- and low-confidence hits were directly compared (Table 6).